The emergence of silicon MEMS brought radical improvements in the once staid timing industry and now MEMS timing solutions are rapidly replacing quartz products.
By Piyush Sevalia, Executive VP of Marketing & Dr. Aaron Partridge, Chief Scientist, SiTime.
Timing devices provide the clock signal or heartbeat in digital electronic systems to which all other signals are synchronised. Over the past several decades, these timing devices have been based on quartz crystals, available in the form of resonators (a mechanical vibrating element) and oscillators (a resonator combined with an electrical circuit).
Quartz crystals are built in specialised factories by companies that are highly focused to this task. The core competency of these quartz companies lies in the precision manufacturing and cutting of quartz crystals to ensure they operate at the correct frequency and are stable over temperature. Quartz manufacturers have done an excellent job in fulfilling these basic and essential needs. However, quartz companies have not kept pace with the semiconductor industry in terms of performance improvement and features, nor are quartz companies well equipped to provide the variety, fast turnaround and cost advantages inherent in the semiconductor industry.
Quartz crystal oscillators are limited in accuracy to about ±20 ppm for non-compensated devices. Frequencies are limited to a range of 10MHz to 60MHz for small packages. Jitter is in the range of 1 to 2ps, integrated over 12kHz to 20MHz. These limitations are set by the mechanical constraints of quartz devices. While these limits can be exceeded in special cases, the cost increases significantly. MEMS oscillators do not have these limits since they use a programmable analog architecture. For instance the circuit-centric MEMS oscillator architecture readily supports frequencies from 1 to 625MHz.
There are also subtle problems in quartz that are not found in silicon MEMS. A phenomenon called ‘activity dips’ is a good example. Activity dips set the lower limit on crystal stability at about 0.1ppm, or about 100ppb. They cause a resonator’s frequency to sharply jump tens of ppb as the part is swept over temperature. In low cost crystals it can be much worse, with jumps of 1ppm. The reason for activity dips is related to how waves propagate laterally through quartz; they are extremely difficult to remove and for practical purposes one must assume that all quartz resonators have them. Activity dips are not present in correctly designed silicon MEMS.
Quartz companies typically outsource their oscillator circuit development and production to semiconductor companies and focus their resources on manufacturing the quartz crystals. In contrast, silicon MEMS timing companies follow the fabless semiconductor model and have deep expertise in designing the MEMS resonator as well as the analog oscillator circuit. This in-house intelligence results in the availability of unique features that are not available from quartz oscillators. MEMS timing features include: customisable frequency from 1Hz to 625MHz with up to six decimal places of accuracy; spread spectrum capability for EMI reduction; programmable drive strength control for better signal integrity and EMI reduction; 1.8V operation over the entire frequency range and 1.2 to 4.5V (continuous) operation for battery powered applications, and; programmable pull range from ±25 to ±1600ppm in VCXOs, VCTCXOs and DCXOs
These features are available in MEMS devices over a variety of operating temperatures and a wide range of industry standard SMD packages that can be used as drop-in replacement for quartz devices. Special packages are also available, such as ultra-small 1.5x0.8mm chip-scale packages or SOT23-5 for higher board-level reliability in harsh environments.
Because MEMS oscillators have a programmable architecture, most features can be customised using a programmer such as SiTime’s Time Machine II. This small programmer can create instant samples in any frequency, any stability and any supply voltage. This gives system designers the capability to program and test a vast array of timing-related features in their own lab and accelerate development time without needing to search, source and wait for samples.
Silicon MEMS timing devices are manufactured in semiconductor fabs and assembly houses. MEMS oscillators are held in inventory in the form of programmable die on wafers. When a mass production order is placed, devices are packaged, tested, programmed, and shipped with a three to five week lead-time. This is much shorter than the typical 8-16 week manufacturing lead-time of quartz device makers that follow a material-intense manufacturing flow. The short lead times offered by MEMS timing vendors result in better inventory control, as well as the ability to meet upside demand more quickly and cost-effectively.
Silicon MEMS timing solutions also exhibit better reliability (operating life) compared to quartz. MEMS-based oscillators have an FIT rate of less than two, which translates to 500m hours MTBF. This is approximately 15 times better than typical quartz devices. In terms of robustness and immunity to noise, SiTime MEMS oscillators demonstrate much better electromagnetic susceptibility compared to comparable common quartz-based oscillators, as well as better power supply noise rejection and immunity to shock, to name but a few.
These benefits stem from the size and structure of the resonators. The resonators in quartz oscillators are millimetre-scale cantilevered structures that are sensitive to acceleration. While a crystal may have MHz electrical resonances, it has kHz structural resonances. These kHz frequencies can be excited by external vibration or shock, and this shows up as vibration sensitivity or failures. MEMS resonators on the other hand, are about 10 times smaller with up to 3000 times smaller moving mass and have about 10 times higher mechanical modes and are thus less sensitive to external vibration and shock. It is a matter of simple scaling. Additionally, packaging and oscillator circuit design, such as those used in SiTime oscillators, make MEMS-based devices more immune to electrical noise. For example, the MEMS resonators are mounted close to the drive circuitry, giving less antenna area for electrical noise pickup compared to quartz packaging. Multi-level on-chip regulators make the oscillator more resilient against power supply noise.
Because MEMS devices are manufactured in silicon and packaged in low-cost standard plastic packaging, they also offer a lower price trajectory. MEMS timing companies, which use a fabless model, leverage the infrastructure of the semiconductor industry and are therefore better equipped to offer competitive pricing. In addition, the short lead-times, increased features and higher reliability of MEMS timing solutions translate to a lower total cost of ownership for electronics manufacturers.
SiTime offers kHz and MHz resonators for customer that want to integrate MEMS resonators into their own products. The kHz resonators are suited to timekeeping applications where one would use a 32kHz quartz tuning fork, and the MHz resonators are suitable for reference applications such as clocking and RF. It is difficult to embed quartz crystals inside plastic packages as the over-moulding causes significant performance and reliability issues.
Time for MEMS solutions
Until the past decade, system designers and OEMs have had to work around the limitations of quartz products as it was the only viable option. As MEMS timing solutions have entered the market, the limitations in performance, reliability and flexibility have been lifted. As advances in MEMS timing solutions accelerate, designers are given new capabilities to improve system performance, features, size and power, while supply chain managers are no longer constricted by quartz availability and cost issues. The gap between quartz and MEMS solutions will continue to widen as MEMS companies follow the pace of improvement within the semiconductor industry driven by Moore’s law.
The transition to silicon has been seen repeatedly in other areas. Silicon has replaced film in still cameras and tape in video cameras. It has replaced magnetic disks and vacuum tube displays. Newspapers and books are converting to silicon-based readers. Silicon is driving the closing act of Gutenberg’s revolution in publishing; our grandchildren will not read books printed on paper. Our grandchildren will also not be using devices timed with quartz. For that matter, in a few years we won’t be either.